Quantification of silica transport in subduction zones using machine learning on geochemical data of pelitic rocks

Abstract

Silica is highly soluble in crustal fluids, and its transport and precipitation control the permeability and fluid pressure of faults. In particular, silica sealing in seismogenic zones is thought to influence the recurrence interval of earthquakes, but its spatial distribution remains poorly understood.In this study, we investigated the distribution and depth-dependent variation of silica sealing in the Sanbagawa metamorphic belt by integrating field observations with geochemical mass transfer analysis based on machine learning. Using 95 metapelite samples collected across a temperature range of 288-536 degrees C, we quantified silica mobility by reconstructing protolith compositions with the Protolith Reconstruction Model (PRM).Our results revealed significant silica addition (+32 wt%) at an outcrop corresponding to 300 degrees C, while high-grade metamorphic zones exhibited silica depletion (up to -9 wt%). The silica addition coincides with the transition between slow and fast earthquake zones, suggesting that silica sealing may contribute to overpressure development at depth.